Symbol Tables

ASU Textbook Chapter 7.6, 6.5 and 6.3

Tsan-sheng Hsu
tshsu@iis.sinica.edu.tw

http://www.iis.sinica.edu.tw/~tshsu
1

Denitions
Symbol table: A data structure used by a compiler to keep track of semantics of variables.
Data type. When is used:

scope.

The eective context where a name is valid.

Where it is stored: storage address.

Possible implementations:
Unordered list: for a very small set of variables. Ordered linear list: insertion is expensive, but implementation is

relatively easy. Binary search tree: O(log n) time per operation for n variables. Hash table: most commonly used, and very ecient provided the memory space is adequately larger than the number of variables.

Compiler notes #5, Tsan-sheng Hsu, IIS

Hash Table
Hash function h(n): returns a value from 0, . . . , m 1, where n is the input name and m is the hash table size.
Uniform and randomized.

Many design for h(n).

Add up the integer values of characters in a name and then take the

remainder of it divided by m. Add up a linear combination of integer values of characters in a name, and then

Resolving collisions:
Linear resolution: try (h(n) + 1) mod m for m being a prime number. Chaining.
Open hashing. Keep a chain on the items with the same hash value. Most popular.

Quadratic-rehashing:

try (h(n) + 12) mod m, and then try (h(n) + 22) mod m, . . . , try (h(n) + i2) mod m.

Compiler notes #5, Tsan-sheng Hsu, IIS

Performance of Hash Table

Performance issues on using dierent collision resolution schemes. Hash table size must be adequately larger than the maximum number of possible entries. Frequently used variables should be distinct.
Keywords or reserved words. Short names, e.g., i, j and k. Frequently used identiers, e.g., main.

Uniformly distributed.

Compiler notes #5, Tsan-sheng Hsu, IIS

Contents in symbol tables

Possible entries in a symbol table:
Name: a string. Attribute: Reserved word Variable name Type name Procedure name Constant name

Data type. Scope information: where it can be used. Storage allocation, size, . . .

Compiler notes #5, Tsan-sheng Hsu, IIS

How to store names

Fixed-length name: allocated. allocate a xed space for each name

Too little: names must be short. Too much: waste a lot of spaces.

NAME a r

ATTRIBUTES a y

s a r i

o e 2

r a

t d r

Variable-length name:
A string of space is used to store all names. For each name, store the length and starting index of each name.
NAME index length 0 5 5 2 7 10 17 3 ATTRIBUTES

0 s

1 o

2 r

3 t

4 $

5 a

6 $

7 r

8 e

9 a

10 d

11 a

12 r

13 r

14 a

15 y

16 $

17 i

18 2

19 $

Compiler notes #5, Tsan-sheng Hsu, IIS

Handling block-structures
Nested block means nested scope. Example (C language code)
main() { /* open a new scope */ int H,A,L; /* parse point A */ ... { /* open another new scope */ float x,y,H; /* parse point B */ ... /* x and y can only be used here */ /* H used here is float */ ... } /* close an old scope */ ... /* H used here is integer */ ... { char A,C,M; /* parse point C */ ... } }

Compiler notes #5, Tsan-sheng Hsu, IIS

Two common approaches (1/3)

An individual symbol table for each scope.

Use a stack to maintain the current scope. Search top of stack rst. If not found, search the next one in the stack. Use the rst match. Note: a popped scope can be destroyed in a one pass compiler, but it must be saved in a multi-pass compiler.
main() { /* open a new scope */ int H,A,L; /* parse point A */ ... { /* open another new scope */ float x,y,H; /* parse point B */ ... /* x and y can only be used here */ /* H used here is float */ ... } /* close an old scope */ S.T. for ... H, A, L /* H used here is integer */ ... { char A,C,M; /* parse point C */ ... parse point A } }

searching direction

S.T. for x,y,H S.T. for H, A, L

S.T. for A,C,M S.T. for H, A, L

parse point B

parse point C

Compiler notes #5, Tsan-sheng Hsu, IIS

Two common approaches (2/3)

A single global table marked with the scope information.
Each scope is given a unique scope number. Incorporate the scope number into the symbol table.

Two possible codings (among others):

Hash table with chaining. Same names hash into the same location by adding at the front. main() When a scope is closed, all entries of that scope are removed.
{ /* open a new scope */ int H,A,L; /* parse point A */ ... { /* open another new scope */ float x,y,H; /* parse point B */ ... /* x and y can only be used here */ /* H used here is float */ ... } /* close an old scope */ ... /* H used here is integer */ ... { char A,C,M; /* parse point C */ ... }
H(2) L(1) x(2) y(2) A(1) symbol table: hash with chaining parse point B parse point C H(1) H(1) L(1) C(3) M(3) A(3) A(1)

} Compiler notes #5, Tsan-sheng Hsu, IIS 9

Two common approaches (3/3)

A second coding choice:
main() { /* open a new scope */ int H,A,L; /* parse point A */ H(2) ... { /* open another new scope */ float x,y,H; /* parse point B */ ... A(1) /* x and y can only be used here */ /* H used here is float */ ... } /* close an old scope */ ... /* H used here is integer */ ... { char A,C,M; /* parse point C */ parse ... } }

Binary search tree:

H(1) L(1) x(2) y(2) point B A(1) A(3)

H(1) L(1) C(3) M(3)

parse point C

It is dicult to close a scope.

Need to maintain a list of entries in the same scope. Using this list to close a scope and to reactive it for the second pass.

Compiler notes #5, Tsan-sheng Hsu, IIS

10

Records and elds

The with construct in PASCAL can be considered an additional scope rule.
Field names are visible in the scope that surrounds the record declara-

tion. Field names need only to be unique within the record.

Example (PASCAL code): A, R: record A: integer X: record A: real; C: boolean; end end ... R.A := 3; /* means R.A := 3; */ with R do A := 4; /* means R.A := 4; */
Compiler notes #5, Tsan-sheng Hsu, IIS 11

Implementation of eld names

Two choices for handling eld names:
Allocate a symbol table for each record type used.
main symbol table another symbol table A A record X R record another symbol table A another symbol table A integer X another symbol table A real record C boolean real record integer

boolean

Associate a record number within the eld names. Assign record number #0 to names that are not in records. A bit time consuming in searching the symbol table. Similar to the scope numbering technique.
Compiler notes #5, Tsan-sheng Hsu, IIS 12

Implementation of PASCAL with construct

Example: with R do begin A := 3; with X do A := 3.3 end If each record (each scope) has its own symbol table,
then push the symbol table for the record onto the STACK.

If the record number technique is used,

then keep a stack containing the current record number during searching, success only if it matches the current number. If fail, then use next record number in the stack as the current record

number and continue to search. If everything fails, search the normal main symbol table.

Compiler notes #5, Tsan-sheng Hsu, IIS

13

Overloading (1/3)
A symbol may, depending on context, mean more than one thing. Example:
operators: I := I + 3; X := Y + 1.2; function call return value and recursive function call: f := f + 1;

Compiler notes #5, Tsan-sheng Hsu, IIS

14

Overloading (2/3)
Implementation:
Link together all possible denitions of an overloading name. Call this an

overloading chain. Whenever a name that can be overloaded is dened

if the name is already in the current scope, then add the new denition in the overloading chain; if it is not already there, then enter the name in the current scope, and link the new entry to any existing denitions; search the chain for an appropriate one, depending on the context.

Whenever a scope is closed, delete the overloading denitions from

the head of the chain.

Compiler notes #5, Tsan-sheng Hsu, IIS

15

Overloading (3/3)
Example: PASCAL function name and return variable.
Within the function body, the two denitions are chained. i.e., function call and return variable. When the function body is closed, the return variable denition disap-

pears.

[PASCAL] function f: integer; begin if global > 1 then f := f +1; return end

Compiler notes #5, Tsan-sheng Hsu, IIS

16

Forward reference (1/2)

Denition:
A name that is used before its denition is given. To allow mutually referenced and linked data types, names can some-

times be used before it is declared.

GOTO labels:
If labels must be dened before its usage, then one-pass compiler

suces. Otherwise, we need either multi-pass compiler or one with backpatching.

Avoid resolving a symbol until all its possible denitions have been seen. In C, ADA and languages commonly used today, the scope of a declaration extends only from the point of declaration to the end of the containing scope.

Compiler notes #5, Tsan-sheng Hsu, IIS

17

Forward reference (2/2)

Pointer types:
determine the element type if possible; chaining together all references to a pointer to type T until the end of

the type declaration; all type names can then be looked up and resolved. [PASCAL] type link = ^ cell; cell = record info: integer; next: link; end;

Compiler notes #5, Tsan-sheng Hsu, IIS

18

Type equivalent and others

How to determine whether two types are equivalent?

Structural equivalence:
Express a type denitions using a directed graph using nodes as entries. Two types are equivalent if and only if their structures (graphs) are the same. A dicult job for compilers.

Name equivalence:
Two types are equivalent if and only if their names are the same. An easy job for compilers, but the coding takes more time.

Symbol table is needed during compilation, might also be needed during debugging.

Compiler notes #5, Tsan-sheng Hsu, IIS

19

How to use?
Dene symbol tbale routines:
Find in symbol table(name,scope):
return not found or an entry in the symbol table

check whether a name within a particular scope is currently in the symbol table or not.

Insert into sumbol table(name,scope) Return the newly created entry. Delete from sumbol table(name,scope)

Grammar productions:
Declaration: D T L { insert each name in $2.namelist into symbol table, allocate sizeof ($1.type) bytes, error for duplicated names} T int{$$.type = int} L id, L {insert the new name into $3.namelist and put it in $$.namelist} | id {create a list of one name $$.namelist} Allocate global and temperatory data space at the end of code. P program end {printf(GDATA:\n); printf(nbytes %d\n,total Gsize); printf(TDATA:\n); printf(nbytes %d\n,total Tsize); }
Compiler notes #5, Tsan-sheng Hsu, IIS 20

More issues on usage

Expressions:
S E + E { generate code for adding data at $1.taddr and $3.taddr} printf(load R1,TDATA+%d\n,$1.taddr); printf(load R2,TDATA+%d\n,$3.taddr); free $1.taddr and $3.taddr from temp space; printf(add R1,R2\n); current t =allocate temp space; printf(store TDATA+%d, R1\n,current t); $$.taddr = current t E id {nd symbol table entry, allocate at global adta space gadd} printf(load R1, GDATA+%d\n,gaddr ); current t =allocate temp space; printf(store TDATA+%d, R1\n,current t); $$.taddr = current t

Compiler notes #5, Tsan-sheng Hsu, IIS

21